A Data Scientist Used ChatGPT and AlphaFold to Build a Cancer Vaccine for His Dog

Paul Conyngham is a Sydney-based data analyst with 17 years of machine learning experience and zero background in biology. When veterinarians gave his rescue dog Rosie one to six months to live after surgery and chemotherapy failed, he did what any desperate engineer would do — he asked ChatGPT for help.

The Story

Rosie, a Staffy-Shar Pei cross, had mast cell tumors — the most common skin cancer in dogs. A tennis ball-sized growth on her leg. Standard treatments had failed. Conyngham turned to AI not as a magic cure, but as a research navigator.

ChatGPT mapped out the plan: get genomic sequencing done (one healthy sample, one tumor), identify the mutations, design a vaccine targeting them. It pointed him to the UNSW Ramaciotti Centre for Genomics, recommended specific equipment, and even suggested which researcher to contact — Dr. Martin Smith, a computational biologist who agreed to run the sequencing.

The $3,000 sequencing produced 320 gigabytes of raw data. Conyngham used ChatGPT to sift through it and identify neoantigens — mutated proteins on the tumor surface. He then used Google DeepMind's AlphaFold to model Rosie's mutated c-KIT protein against the healthy baseline.

What happened next surprised even the researchers involved. Prof. Pall Thordarson at the UNSW RNA Institute — who had produced hundreds of different mRNA molecules but never a cancer vaccine — took Conyngham's design and manufactured the actual vaccine, packaging it in lipid nanoparticles for delivery.

After a three-month ethics approval process and a 100-page application, Rosie received her first injection in combination with an immune checkpoint inhibitor.

The tumors shrank by roughly 75%. Within a month, Rosie went from limited mobility to jumping over a fence to chase a rabbit.

The Important Caveats

This is not a cure. Prof. Thordarson was candid: "This may not have cured Rosie. Bought time for sure, yes, but some of the tumours didn't respond." The vaccine was administered alongside a checkpoint inhibitor, making it difficult to separate which treatment had the greater effect. The AlphaFold protein model carried a confidence score of just 54.55 — which UNSW structural biologist Dr. Kate Michie described as "low."

Conyngham himself acknowledged the limits: "I'm under no illusion that this is a cure, but I do believe this treatment has bought Rosie significantly more time and quality of life."

And the final vaccine construct, as Conyngham revealed, was actually designed by Grok, with Gemini handling much of the computational work. ChatGPT's role was primarily as a research navigator and data explorer. The heavy scientific lifting — sequencing, mRNA production, formulation, safety protocols — was done by human researchers at UNSW.

Why It Matters Anyway

The story went viral after OpenAI co-founder Greg Brockman and DeepMind chief Demis Hassabis both amplified it. Steven Hsesheng Lin at MD Anderson Cancer Center called the concept-to-treatment timeline "astounding."

What's genuinely remarkable isn't that AI cured cancer — it didn't. It's that a person with no biomedical training used freely available AI tools to navigate a complex scientific domain, identify the right experts, and contribute meaningfully to a treatment that extended his dog's life. The distance between "I have a problem" and "I found the right people and data to address it" collapsed from years to weeks.

A Phase III clinical trial is already exploring personalized mRNA vaccines with checkpoint inhibitors for human lung cancer. Rosie's case, imperfect as it is, hints at a future where that kind of personalization becomes accessible far beyond elite research hospitals.